Subject: | |
From: | |
Reply To: | |
Date: | Wed, 25 Apr 2007 08:34:19 -0500 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Title: Understanding Wordscores
Authors: Will Lowe
Entrydate: 2007-04-25 07:34:50
Keywords: content analysis, wordscores, ideal point, item response
theory
Abstract: Wordscores (Laver, Benoit & Garry, 2003) is a procedure
for inferring scores for new documents on the basis of scores for
words derived from previously scored documents. It has important
virtues - it is computationally straightforward, makes no
distributional or functional assumptions, and scores on a specific
dimension (it is an a priori, not inductive method), but also some
concerning vices - estimated document scores are on the wrong scale,
there are no explicit distributional or functional assumptions, and
there is no explanation of how Wordscores is an a priori method.
This paper investigates why document scores are on the wrong scale
and asks when Wordscores work well, what probabilistic model would
motivate the procedure, and what makes an a document scoring method a
priori. In answering these questions we show that that document score
estimation implies problematic linguistic assumptions, word score
estimation is biased and inconsistent, and that Wordscores can be
motivated by an ideal point form of latent variable model for words
having the reduced form of a Rhetorical Ideal Point / Wordfish model.
http://polmeth.wustl.edu/retrieve.php?id=691
|
|
|